Merge branch 'main' into update_crend (3f287a66) · Commits · SA4 / Audio / IVAS BASOP

.gitlab-ci.yml

+1 −1

Original line number	Diff line number	Diff line
		@@ -654,7 +654,7 @@ build-codec-linux-make:
		rules:
		- if: $CI_PIPELINE_SOURCE == 'web'
		- if: $CI_PIPELINE_SOURCE == 'push' && $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH
		- if: $CI_PIPELINE_SOURCE == 'merge_request_event' && $CI_MERGE_REQUEST_TARGET_BRANCH_NAME == "main" # only have MR pipelines for MRs to main
		- if: $CI_PIPELINE_SOURCE == 'merge_request_event' # trigger build job for all MRs
		- if: $CI_PIPELINE_SOURCE == 'schedule'
		- if: $CI_PIPELINE_SOURCE == 'push'
		when: never

lib_com/cldfb_evs.c

+4 −0

Original line number	Diff line number	Diff line
		@@ -1405,7 +1405,11 @@ void GetEnergyCldfb( Word32 energyLookahead, /!< o: Q(*sf_energyLookahead)
		{
		FOR( j = 20; j < numberBandsM; j++ )
		{
		#ifdef BASOP_NOGLOB
		nrg = L_add_o( nrg, L_shr_o( energyValues[k][j], s, &Overflow ), &Overflow );
		#else
		nrg = L_add( nrg, L_shr( energyValues[k][j], s ) );
		#endif
		}
		}

lib_com/enhancer_fx.c

+18 −34

Original line number	Diff line number	Diff line
		@@ -600,9 +600,11 @@ void enhancer_ivas_fx2(
		Word16 tmp, fac, *pt_exc2;
		Word16 i;
		Word32 L_tmp;
		Word32 L_tmp1, L_tmp2;
		Word16 gain_code_hi;
		Word16 pit_sharp, tmp16;
		Word16 excp[L_SUBFR], sc;
		Word64 w_temp;


		pit_sharp = gain_pit;
		@@ -751,50 +753,32 @@ void enhancer_ivas_fx2(
		* filter with coefs [-tmp 1.0 -tmp] where tmp=0...0.25.
		* This is applied to code and add_fxed to exc2
		-----------------------------------------------------------------/

		L_tmp1 = L_deposit_h( gain_code_hi ); // Q = 16 + Q_exc, gain_code
		L_tmp2 = L_mult( tmp, gain_code_hi ); // Q = 16 + Q_exc, gain_code * temp

		/* pt_exc2[0] += code[0] - tmp * code[1] */
		L_tmp = L_deposit_h( code[0] ); /* if Enc :Q9 * Q15 -> Q25 */
		L_tmp = L_msu( L_tmp, code[1], tmp ); /* Q12 * Q15 -> Q28 */
		#ifdef BASOP_NOGLOB
		L_tmp = L_shl_sat( L_mult( gain_code_hi, extract_h( L_tmp ) ), sc );
		pt_exc2[0] = msu_r_sat( L_tmp, -32768, pt_exc2[0] );
		w_temp = W_msu_32_16( W_mult_32_16( L_tmp1, code[0] ), L_tmp2, code[1] ); // Q = 32 + Q_exc - sc (16+Q_exc+15-sc+1)
		w_temp = W_shl( w_temp, sc ); // Q = 32 + Q_exc
		pt_exc2[0] = W_round32_s( W_msu_32_16( w_temp, MIN_32, pt_exc2[0] ) ); // Q = Q_exc
		move16();
		#else
		L_tmp = L_shl( L_mult( gain_code_hi, extract_h( L_tmp ) ), sc );
		pt_exc2[0] = msu_r( L_tmp, -32768, pt_exc2[0] );
		move16();
		#endif
		move16(); /* in Q_exc */

		FOR( i = 1; i < L_SUBFR - 1; i++ )
		{
		/* pt_exc2[i] += code[i] - tmp * code[i-1] - tmp * code[i+1] */
		L_tmp = L_msu( -32768, code[i], -32768 );
		L_tmp = L_msu( L_tmp, code[i + 1], tmp );
		#ifdef BASOP_NOGLOB
		tmp16 = msu_r_sat( L_tmp, code[i - 1], tmp );
		L_tmp = L_shl_sat( L_mult( gain_code_hi, tmp16 ), sc );
		pt_exc2[i] = msu_r_sat( L_tmp, -32768, pt_exc2[i] );
		/* pt_exc2[i] += (code[i] - tmp * code[i-1] - tmp * code[i+1]) * gain_code */
		w_temp = W_msu_32_16( W_mult_32_16( L_tmp1, code[i] ), L_tmp2, code[i - 1] ); // Q = 32 + Q_exc - sc (16+Q_exc+15-sc+1)
		w_temp = W_msu_32_16( w_temp, L_tmp2, code[i + 1] ); // Q = 32 + Q_exc - sc (16+Q_exc+15-sc+1)
		w_temp = W_shl( w_temp, sc ); // Q = 32 + Q_exc
		pt_exc2[i] = W_round32_s( W_msu_32_16( w_temp, MIN_32, pt_exc2[i] ) ); // Q = Q_exc
		move16();
		#else
		tmp16 = msu_r( L_tmp, code[i - 1], tmp );
		L_tmp = L_shl( L_mult( gain_code_hi, tmp16 ), sc );
		pt_exc2[i] = msu_r( L_tmp, -32768, pt_exc2[i] );
		#endif
		move16(); /* in Q_exc */
		}

		/* pt_exc2[L_SUBFR-1] += code[L_SUBFR-1] - tmp * code[L_SUBFR-2] */
		L_tmp = L_deposit_h( code[L_SUBFR - 1] ); /Q28 /
		L_tmp = L_msu( L_tmp, code[L_SUBFR - 2], tmp ); /Q28 /
		L_tmp = L_shl( L_mult( gain_code_hi, extract_h( L_tmp ) ), sc );
		#ifdef BASOP_NOGLOB
		pt_exc2[L_SUBFR - 1] = msu_r_sat( L_tmp, -32768, pt_exc2[L_SUBFR - 1] );
		w_temp = W_msu_32_16( W_mult_32_16( L_tmp1, code[L_SUBFR - 1] ), L_tmp2, code[L_SUBFR - 2] ); // Q = 32 + Q_exc - sc (16+Q_exc+15-sc+1)
		w_temp = W_shl( w_temp, sc ); // Q = 32 + Q_exc
		pt_exc2[L_SUBFR - 1] = W_round32_s( W_msu_32_16( w_temp, MIN_32, pt_exc2[L_SUBFR - 1] ) ); // Q = Q_exc
		move16();
		#else
		pt_exc2[L_SUBFR - 1] = msu_r( L_tmp, -32768, pt_exc2[L_SUBFR - 1] );
		move16();
		#endif
		move16(); /* in Q_exc */

		test();
		test();
		IF( Opt_AMR_WB && ( EQ_32( core_brate, ACELP_8k85 ) \|\| EQ_32( core_brate, ACELP_6k60 ) ) )

lib_com/int_lsp.c

+1 −0

Original line number	Diff line number	Diff line
		@@ -344,6 +344,7 @@ void int_lsp4_ivas_fx(
		}
		ELSE IF( EQ_16( relax_prev_lsf_interp, -2 ) )
		{
		// This condition not present in int_lsp4_fx
		pt_int_coeffs = interpol_frac2_mid_fx; /Q15/
		}
		ELSE

lib_com/ivas_prot.h

+48 −47

Original line number	Diff line number	Diff line
		@@ -229,49 +229,50 @@ ivas_error pre_proc_front_ivas(
		ivas_error pre_proc_front_ivas_fx(
		SCE_ENC_HANDLE hSCE, /* i/o: SCE encoder structure */
		CPE_ENC_HANDLE hCPE, /* i/o: CPE encoder structure */
		const int32_t element_brate, /* i : SCE/CPE element bitrate */
		const int16_t nb_bits_metadata, /* i : number of metadata bits */
		const int16_t input_frame, /* i : frame length */
		const int16_t n, /* i : channel number */
		Word16 old_inp_12k8_fx[], /* o : buffer of old input signal */
		const Word32 element_brate, /* i : SCE/CPE element bitrate */
		const Word16 nb_bits_metadata, /* i : number of metadata bits */
		const Word16 input_frame, /* i : frame length */
		const Word16 n, /* i : channel number */
		Word16 old_inp_12k8_fx[], /* o : (Q-1) buffer of old input signal */
		float old_inp_12k8[], /* o : buffer of old input signal */
		float old_inp_16k[], /* o : buffer of old input signal @16kHz */
		float ener, / o : residual energy from Levinson-Durbin */
		float relE, / o : frame relative energy */
		float A[NB_SUBFR16k * ( M + 1 )], /* o : A(z) unquantized for the 4 subframes */
		float Aw[NB_SUBFR16k * ( M + 1 )], /* o : weighted A(z) unquantized for subframes */
		Word16 relE_fx, / o : frame relative energy Q8 */
		Word16 A_fx[NB_SUBFR16k * ( M + 1 )], /* o : A(z) unquantized for the 4 subframes */
		Word16 Aw_fx[NB_SUBFR16k * ( M + 1 )], /* o : weighted A(z) unquantized for subframes */
		float epsP[M + 1], /* o : LP prediction errors */
		float lsp_new[M], /* o : LSPs at the end of the frame */
		float lsp_mid[M], /* o : LSPs in the middle of the frame */
		int16_t vad_hover_flag, / o : VAD hangover flag */
		int16_t attack_flag, / o : flag signaling attack */
		float realBuffer[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX], /* i/o: real buffer */
		float imagBuffer[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX], /* i/o: imag buffer */
		Word16 lsp_new_fx[M], /* o : LSPs at the end of the frame Q15 */
		Word16 lsp_mid_fx[M], /* o : LSPs in the middle of the frame Q15 */
		Word16 vad_hover_flag, / o : VAD hangover flag */
		Word16 attack_flag, / o : flag signaling attack */
		Word32 realBuffer_fx[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX], /* i/o: real buffer Q(q_re_im_buf) */
		Word32 imagBuffer_fx[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX], /* i/o: imag buffer Q(q_re_im_buf) */
		Word16 q_re_im_buf, / i/o: Q-factor of real and imag buffer */
		float old_wsp[], /* o : weighted input signal buffer */
		float pitch_fr[NB_SUBFR], /* o : fractional pitch values */
		float voicing_fr[NB_SUBFR], /* o : fractional pitch gains */
		int16_t loc_harm, / o : harmonicity flag */
		float cor_map_sum, / o : speech/music clasif. parameter */
		int16_t vad_flag_dtx, / o : HE-SAD flag with additional DTX HO */
		Word16 pitch_fr_fx[NB_SUBFR], /* o : fractional pitch values Q6 */
		Word16 voicing_fr_fx[NB_SUBFR], /* o : fractional pitch gains Q15 */
		Word16 loc_harm, / o : harmonicity flag */
		Word16 cor_map_sum_fx, / o : speech/music clasif. parameter Q8 */
		Word16 vad_flag_dtx, / o : HE-SAD flag with additional DTX HO */
		float enerBuffer[CLDFB_NO_CHANNELS_MAX], /* o : energy buffer */
		float fft_buff[2 * L_FFT], /* o : FFT buffer */
		Word16 fft_buff_fx[2 * L_FFT], /* o : FFT buffer */
		const float tdm_A_PCh[M + 1], /* i : unq. LP coeff. of primary channel */
		const float tdm_lsp_new_PCh[M], /* i : unq. LSPs of primary channel */
		const float currFlatness, /* i : flatness parameter */
		const int16_t tdm_ratio_idx, /* i : Current Ratio_L index */
		const Word16 tdm_A_PCh_fx[M + 1], /* i : unq. LP coeff. of primary channel Q12 */
		const Word16 tdm_lsp_new_PCh_fx[M], /* i : unq. LSPs of primary channel Q15 */
		const Word16 currFlatness_fx, /* i : flatness parameter Q7 */
		const Word16 tdm_ratio_idx, /* i : Current Ratio_L index */
		float fr_bands_LR[][2 * NB_BANDS], /* i : energy in frequency bands */
		const float Etot_LR[], /* i : total energy Left & Right channel */
		const Word16 Etot_LR_fx[], /* i : total energy Left & Right channel Q8*/
		float lf_E_LR[][2 * VOIC_BINS], /* i : per bin spectrum energy in lf, LR channels */
		const int16_t localVAD_HE_SAD_LR[], /* i : HE-SAD flag without hangover, LR channels */
		const Word16 localVAD_HE_SAD_LR[], /* i : HE-SAD flag without hangover, LR channels */
		float band_energies_LR[2 * NB_BANDS], /* o : energy in critical bands without minimum noise floor E_MIN */
		const int16_t flag_16k_smc, /* i : flag to indicate if the OL SMC is run at 16 kHz */
		const int16_t front_vad_flag, /* i : front-VAD flag to overwrite VAD decision */
		const int16_t force_front_vad, /* i : flag to force VAD decision */
		const int16_t front_vad_dtx_flag, /* i : front-VAD DTX flag to overwrite VAD decision*/
		const Word16 flag_16k_smc, /* i : flag to indicate if the OL SMC is run at 16 kHz */
		const Word16 front_vad_flag, /* i : front-VAD flag to overwrite VAD decision */
		const Word16 force_front_vad, /* i : flag to force VAD decision */
		const Word16 front_vad_dtx_flag, /* i : front-VAD DTX flag to overwrite VAD decision*/
		const IVAS_FORMAT ivas_format, /* i : IVAS format */
		const int16_t MCT_flag, /* i : hMCT handle allocated (1) or not (0) */
		const int32_t ivas_total_brate /* i : IVAS total bitrate */
		const Word16 MCT_flag, /* i : hMCT handle allocated (1) or not (0) */
		const Word32 ivas_total_brate /* i : IVAS total bitrate - for setting the DTX */
		);
		ivas_error pre_proc_ivas(
		Encoder_State st, / i/o: encoder state structure */
		@@ -611,6 +612,7 @@ void stereo_tcx_core_dec(
		);
		#endif // !IVAS_FLOAT_FIXED

		#ifndef IVAS_FLOAT_FIXED
		void stereo_tcx_core_enc(
		Encoder_State st, / i/o: encoder state structure */
		const float new_samples_12k8[], /* i : buffer of input signal @12.8 kHz */
		@@ -622,6 +624,7 @@ void stereo_tcx_core_enc(
		const Word16 last_element_mode, /* i : last element mode, Q0 */
		const Word16 vad_hover_flag /* i : VAD hangover flag, Q0 */
		);
		#endif


		void stereo_tcx_init_dec(
		@@ -2276,7 +2279,6 @@ void td_stereo_param_updt(
		const int16_t flag_ACELP16k, /* i : ACELP@16kHz flag */
		const int16_t tdm_use_IAWB_Ave_lpc /* i : flag to indicate the usage of mean inactive LP coefficients */
		);
		#endif

		void gsc_enc(
		Encoder_State st, / i/o: State structure */
		@@ -2290,7 +2292,6 @@ void gsc_enc(
		float tmp_noise / o : long-term noise energy */
		);

		#ifndef IVAS_FLOAT_FIXED
		void tdm_low_rate_enc(
		Encoder_State st, / i/o: State structure */
		const float Aq[], /* i : 12k8 Lp coefficient */